Modeling and Experimental Comparison of Effects of a Forging Post-Process on Additive Manufactured Metallurgy

Timothy Cyders, Ph.D.

Assistant Professor of Mechanical Engineering

Russ College of Engineering and Technology, Ohio University

Students involved in project:

Cody Mackey, Graduate Student, Mechanical Engineering

Joseph Juratovac, Graduate Student, Mechanical Engineering

Claudia Prieto, Graduate Student, Chemical and Biomolecular Engineering

Megan Clum, Undergraduate Student, Mechanical Engineering

Abstract This study examines the effect of post-process compressive plastic deformation on the mechanical

performance of metal made by direct metal laser sintering (DMLS), an additive process. 316L stainless

steel samples made in standard sheet form were compared to DMLS samples of the same geometry in

both an as-manufactured state and a rolled state in regards to porosity, tensile performance, fatigue

performance, and corrosion resistance. Deformation was introduced through a cold rolling process to

avoid confounding of thermal effects. Plastically deformed DMLS material exhibited lower porosity than

as-manufactured samples, and their tensile performance was enhanced beyond what was explained

through densification and conventional modeling of strain hardening and redundant work. Fatigue

performance of DMLS material was far lower than that of sheet samples, and was unaffected by plastic

deformation. Plastic deformation appeared to have a significant effect on mass loss in a pitting corrosion

attack on the DMLS material, reducing the loss rate to that exhibited in traditionally manufactured

material. It appears from these results that forging and related deformation processes could serve as an

important post-process to maximize the tensile and corrosion performance of DMLS metals, but more

characterization would be needed to understand the phenomena that dominate DMLS material fatigue

performance and how such post-processes could be employed to effectively improve that performance

as well.

Acknowledgements

The author and all the students who were involved in this project would like to thank the Forging

Industry Educational and Research Foundation for funding this work, as well as the involvement of

contacts through FIERF in industry who informed and improved the project with their experience and

knowledge.

1. Introduction

Additive manufacturing (AM) has become a specific topic of interest in the general manufacturing

community due to its novelty and projected abilities for ground-up automation. In metalworking, AM

processes have been studied in various depths over the past 20 years, just recently making significant

strides as the popularity of metal processes grows. Despite this growth, understanding of the effects of

AM artifacts on mechanical behavior is relatively basic. Most metal-based AM processes have required

significant post-processing to account for deficiencies in mechanical performance; most mainstream

material treatments are incapable of working the material, and thus cannot take advantage of the

superior performance imparted by deformation-based processes. From a process perspective, it is likely

that AM will be most attractive in conjunction or combination with other traditional processes, which

may include forging, if the mechanical behavior of metals with AM artifacts can be understood

sufficiently to appropriately design such a process.

In short, AM has two advantages: highly specific geometries can be made with generally the same

ease as simple geometries, and internal voids and passageways can be easily built into the part allowing

for easy lightweighting and geometric optimization. From a design perspective, however, AM has

significant drawbacks in cost and mechanical performance. Part cost in most AM processes is

comparatively high and is tied directly to part volume and print time. Mechanical characteristics such as

strength, isotropy, and fatigue performance usually suffer because of porosity in the layup, layup

directionality, surface finish, and microstructure [1]–[5]. For example, Figure 1 shows a fatigue sample

from this study produced by direct metal laser sintering (DMLS). The print lines are easily visible on the

part as-manufactured. Internally, artifacts known as “scan tracks” are easily visible even after polishing,

and have significant effects on overall porosity, pore geometry, and microstructure. Figure 2 shows a

cross-section of DMLS material etched to show scan tracks and related pores from the welding process.

Compounding these problems is that the various methods to improve mechanical performance

generally slow down the process [6], adding to the cost.

Figure 1: DMLS 316L fatigue sample showing surface finish and visible manufacturing artifacts under a raking light

Figure 2: Scan tracks and pores in DMLS 316L stainless steel etched for 3 seconds with Marble's reagent.

The use of forging as a post-process to AM could provide several significant benefits. Compressive

plastic deformation on the outer layer of an AM preform could reduce porosity (thus improving

mechanical performance), while simultaneously providing better microstructure and surface finish than

current post-processes provide. Isostatic pressing, both cold and hot, is currently used to improve AM

part performance by collapsing microscopic internal voids [7]–[10]. These processes improve material

performance through densification, but do not improve microstructure beyond that of typical, annealed

material.

The core benefit AM can give to forged part design is its ability to produce optimized internal voids.

While microscopic pores generally decrease mechanical performance, macroscopic internal voids such

as those found in trabecular bone could not only mitigate fatigue crack growth, but provide for selective

stiffness, density, and even heat transfer. Open-cell internal structures can be supported during

compressive deformation in a variety of ways, maintaining the functional structures in the preform,

while allowing the forging process to improve the mechanical characteristics of the outer material. In

short, this process could provide an avenue to produce forged parts that benefit from the advantages of

AM, while improving material performance to a level AM alone cannot otherwise achieve.

A number of different AM processes exist for the fabrication of metal parts; two of the dominant

processes are known as direct metal laser sintering (DMLS) and direct metal laser melting (DMLM).

These processes offer relatively good control over part dimensions and the specific build variables that

control the density of the output part, and thus represent a good starting point for this study.

The selection of material for this study is focused on isolating the experimental variables of interest,

with the goal of demonstrating a principle that can be applied to a wider family of materials. In order to

accurately compare these variables to a control group, there must be material with the same chemistry

available through both a powder AM process and a traditional billet process. Complex microstructures

such as those present in titanium alloys would likely confound results, and most traditional aluminum

alloy chemistries are not commonly available through powder AM processes. 316L stainless steel is

available in both powder AM (DMLS/DMLM) and billet form, and its performance as a straight powder

metal has been characterized and studied with respect to densification and other variables [1], [6], [10]–

[13]. It can withstand very high levels of plastic deformation, allowing for this primary variable of

interest to be studied over a wide range of experimental treatments.

2. Objectives

The general objective of this work is the experimental characterization of key mechanical property

improvements (especially porosity, isotropy, strength, and fatigue performance) of simple-geometry AM

materials under varying levels of plastic deformation that would be found in a forging process.

Specifically, this work will address the following hypotheses:

Tensile properties, including yield strength, ultimate tensile strength, and anisotropy of DMLS

SS316L are affected by compressive plastic deformation beyond a level commensurate with

work hardening of the material

High-cycle fatigue life of DMLS SS316L is affected by compressive plastic deformation beyond

what is expected due to simple work hardening of the material

Pitting corrosion resistance of DMLS SS316L is improved by compressive plastic deformation as

compared to samples of the same material with the same surface properties

3. Experimental Setup

The experimental approaches to achieve the objectives include rolling, tensile testing, fatigue, and

corrosion setups and test batteries. Samples were laid up in two primary orientations, as shown in

Figure 3. In the transverse arrangement, material inconsistencies are typically aligned with the stress

area, and therefore usually correlate to lower or more inconsistent tensile performance as compared to

materials laid up in the longitudinal arrangement.

Figure 3: DMLS build orientations compared to stress directionality, with transverse (XY) at left and longitudinal (Z) at right

3.1 Rolling

In order to establish a significant amount of plastic deformation in the samples, rolling was selected

for its ability to easily deform sheet-style samples. The reasoning behind this was two-fold. First and

foremost, the cost of DMLS samples is generally volume-based, so thin sample geometry gives the best

economy for experimentation. Second, this process allowed the samples to be worked cold, such that

plastic deformation could be specifically separated in experiment from the effects of heat. Figure 4

shows the International Rolling Mills model 2040 used to roll the samples in this study.

Figure 4: Rolling mill used to introduce plastic deformation, an International Rolling Mills Lab Rolling Mill Bench Model 2040

Modeling of the rolling process as plane-strain compression is discussed by Hosford and Caddell

[14]. In particular, because this project will involve cold working the material of interest, it is necessary

to address the effects of redundant work in the rolling steps applied to the specimens, while maximizing

the use of the mill’s available load capacity. A basic slab analysis provides that for a reduction r as

described in Equation 1, the effective length of contact, L, between the roll of radius R and the material

is given by Equation 2. In Equation 1, t0 is the initial specimen thickness and tf is the final specimen

thickness. For a roll contact arc of relatively small angle, the average roll pressure, Pav, can be simplified

through a linear assumption to that shown in Equation 3, where σ0 is the average flow stress and tav is

the average thickness between the inlet and outlet of the roll, and µ is the coefficient of friction

between the roll and workpiece. The roll force, F, can then be calculated using the width of the

workpiece, w, as shown in Equation 4.

𝑟 =

𝑡0 − 𝑡𝑓

𝑡0 (1)

𝐿 = √𝑅𝑟ℎ0 (2)

𝑃𝑎𝑣 =

𝑡𝑎𝑣𝜎0𝜇𝐿

(𝑒−𝜇𝐿ℎ − 1) (3)

𝐹 = 𝑃𝑎𝑣𝑤𝐿 (4)

Redundant work, which is not manifest in directly measured shape change, must be modeled in

order to determine whether the difference between the total strain as evidenced by stress-strain

superposition and the apparent strain demonstrated through the measured dimensional change in the

rolling process is significantly different between the DMLS specimens and standard sheet material.

Redundant work is characterized through two parameters, Δ and φ. The Δ parameter is the ratio of

workpiece thickness to contact length, calculated as shown in Equation 5 for flat rolling according to

Hosford and Caddell’s model. Note that this still includes the linearized simplification for the roll contact

area used previously in Equation 3.

Δ =2 − 𝑟

2√𝑡0𝑟𝑅

(5)

Redundant strain, εr, can be calculated by the φ parameter as shown in Equation 6, where εh is the

homogeneous strain. The φ parameter can be further related to the Δ parameter according to Equation

7 for plane strain. This model was applied to the specimen geometry to develop the rolling steps for cold

working the DMLS material, as well as to compare the total strengthening post-process.

ϕ =

𝜖𝑟 + 𝜖ℎ𝜖ℎ

(6)

ϕ = 1 + 0.21(Δ − 1) (7)

Samples were rolled in steps of approximately 5% true strain in the thickness to a maximum of 20%,

which minimized the redundant work in the samples according to Equations 1-7 while maintaining an

appropriate factor of safety against overloading the rolling mill according to Equations 1-4. Rolling was

done cold, and samples were passed through the mill round-robin to maintain consistent adjustments

between rolling steps. The same rolling process was applied to the DMLS and sheet samples for all three

test batteries, so as to maintain a consistent deformation process between test types.

Internal porosity was measured by proxy through density measurements using Archimedes’

Principle. A length of a cut portion of each sample was weighed on an Acculab AL-64 laboratory balance.

Its volume was then determined by water displacement suspended in distilled water by a strand of

monofilament as shown in Figure 5, on the same balance. Samples had an absolute mass of

approximately 3 grams, resulting in maximum relative uncertainties in density of ±0.1%, or about 0.008

g/cc in absolute terms.

Figure 5: Material sample density (and thereby internal porosity) measured by Archimedes' principle

3.2 Tensile Testing

Tensile samples were prepared according to ASTM E8, with the exception that sample geometry was

altered somewhat in order to both maximize gage section space for the two extensometers and to fit

the overall specimens into the available print area with the DMLS manufacturer’s equipment. The

revised specimen geometry is shown in Figure 6. Specimen widths and thicknesses were taken with

calipers and sheet metal micrometers, respectively, and specimens were tested on an Instron 5567 load

frame as shown in Figure 7. Both longitudinal and transverse strain were measured, so as to measure

strain ratio during the tests.

Figure 6: Dimensions of DMLS tensile samples

Figure 7: Tensile specimen loaded, with extensometers, in the load frame

Width extensometer

Length extensometer

Grip

Test specimen

Anisotropy can be characterized by the strain ratio R, not to be confused with the roll radius

previously discussed. For plane strain in sheet metal, this ratio, also referred to as the Lankford

coefficient, describes the ratio of width strain εw to thickness strain εt. Because thickness strain is

difficult to accurately measure for thin sheets, conservation of volume is applied to restate the relation

in terms of width strain and longitudinal strain, εl, as shown in Equation 8. This ratio essentially

represents the sheet material’s resistance to thinning. R values of unity indicate isotropy; according to

Hill’s anisotropic plasticity theory, higher values of R generally indicate increased biaxial tensile strength.

Values for R generally vary somewhat through the plastic region of a tensile test; in this work, readings

were taken at either 15% strain or at the maximum strain before the onset of instability, whichever

came first.

𝑅 =𝜖𝑤𝜖𝑡

=𝜖𝑤

−𝜖𝑙 − 𝜖𝑤 (8)

3.3 Fatigue

As this project used flat samples, a flexure fatigue machine known as a Krouse-type machine was

used to conduct fatigue testing. The ASTM B593 standard describes test procedures and sample designs

specifically for testing of copper sheets, but no such standard exists for a broader set of materials. The

principles laid out in ASTM B593 were applied to 316L stainless steel, and adapted to the expected

modulus values and stress levels for testing in this project. Figure 8 shows a sample mounted in one of

the flexural fatigue testing units at the Center for Advanced Materials Processing (CAMP).

Of specific interest in this project is high cycle fatigue performance, which is generally characterized

through the stress-life or S-N approach. In this approach, a constantly fluctuating stress is applied to a

sample until it forms sufficient cracks for failure. The system described by ASTM B593, as well as the

system depicted in Figure 8, technically exerts a constantly fluctuating strain on the fatigue sample. In

this case, strains and the resulting stresses are generally well within the elastic range of the material,

such that the S-N models correlate well with the calculated values.

Figure 8: Fatigue sample as mounted in flexure fatigue machine, deflected by conrod at right

In this flexural fatigue approach, stress models are specifically related to sample geometry,

deflection, and elastic modulus. In order to accurately characterize stress levels, elastic modulus was

measured for individual samples using the testing setup shown in Figure 9. In this test, calibrated lab

weights are hung from a specially adapted fatigue testing unit with an LVDT to provide high precision

deflection measurements correlated with weight. A linear fit was applied to the weight vs. deflection

data, with typical nonlinearity values less than 2%. This data was then passed to a MATLAB script

conducting a brute-force integration of the sample geometry with a root-finding algorithm to determine

the actual elastic modulus in the flexural direction. This value could then be used in the same MATLAB

script to determine the required deflection for a desired alternating stress level, which could then be

dialed into the fatigue testing machine. This setup generally provided the capability to set the

alternating stress levels to within ± 277 psi with a nominal modulus for steel of 29 Mpsi, and mean stress

levels to within ± 831 psi.

Figure 9: LVDT-based experimental measurement of flexural elastic modulus

Initial stress levels for S-N curve generation were set based on a guess of about 75% of the ultimate

tensile strength from the tensile tests of each material treatment. As additional failure points were

generated, approximate fit curves were applied to better predict failure to create a curve spanning from

the 103-104 cycle regime to the region of expected runout between 106-107 cycles. Because of the high

cost of samples and the high number of samples required, specific staircase testing to establish

endurance limits in addition to S-N curve generation was not possible in the scope of this project. While

most samples were run until fracture, samples that exhibited noticeable cracks without fracturing were

stopped and their modulus was measured; a reduction of modulus greater than 10% was considered to

have failed, and the number of cycles at which the sample was removed was recorded as the failure

point.

Based on typical S-N models, increases in ultimate tensile strength are directly correlated with

alternating stress values for failure at both ends of the S-N curve at 103 and 106 cycles. Austenitic

stainless steels have relatively low notch sensitivity in their annealed state, but high notch sensitivity

when cold worked. Thus, reductions in porosity and additional cold work should demonstrate a

corresponding proportional upward shift of the S-N curve, absent the presence of sharp irregularities.

Significant internal notches coupled with additional cold work, however, would show a corresponding

significant decrease in fatigue performance through a downward shift of the S-N curve. Either way, a

shift of the S-N curve could indicate which phenomena dominate the fatigue performance of the

material.

S-N curves were generated for standard sheet and DMLS 316L specimens in both the as-received

and plastically deformed conditions. Fatigue samples were run through the same rolling schedule as the

tensile samples on a strain basis in order to maintain approximately the same work hardening and

redundant work, such that results could be compared with findings associated with the tensile test

battery. Previous testing has shown that fatigue performance of samples with layer orientation

transverse to the applied load is so poor that it is difficult to even construct an S-N curve. This

experiment was limited to longitudinally oriented samples.

3.4 Corrosion

Because this project used a significant number of samples, it was determined that a corrosion test

could be easily added to the test battery with relatively low cost. A short, comparative study was

conducted through the Institute for Corrosion and Multiphase Technology (ICMT) to determine the

effect of plastic deformation on the corrosion performance of the various material treatments in this

project. For this test, grip sections were cut from the tensile samples after tensile testing was

completed, and were polished to a mirror finish. Three treatments were tested: material type (316L

traditional sheet and DMLS), cold work (0% and 20% rolled), and heat treat (untreated and annealed).

Samples were tested in 10% by weight of FeCl3 in water for 72 hours at 55° C. Resulting pits were

characterized through optical profilometry and scanning electron microscopy, and internal mass

loss/porosity was characterized through Archimedes’ principle as described previously in section 3.1.

4. Results and Discussion

This project was run over the course of the 2016-2017 academic year, and included work from three

graduate students and one undergraduate. Sheet material was purchased through onlinemetals.com

and DMLS specimens were purchased from GPI Prototype and Manufacturing Services.

4.1 Metallurgy

The metallurgy of the 316L DMLS specimens differed significantly from that of the sheet material,

despite having very similar chemistry. Table 1 details the elemental chemistry of the specimens as

measured by direct reading atom emission spectroscopy.

Table 1: Chemical composition of Sheet and DMLS 316L material samples, respectively, from Direct Reading Atom Emission Spectroscopy

Element Al C Cr Cu Mn Mo N Nb Ni P S Si Ti W

Sheet wt.% 0.011 0.01 16.96 0.41 1.23 2.02 0.04 0.014 10.22 0.032 0.006 0.36 0.013 0.05

DMLS wt. % 0.007 0.007 17.61 0.21 1.64 2.68 0.07 0.01 12.1 0.017 0.008 0.26 0.035 0.024

DMLS and sheet samples were mounted, polished and etched with Adler’s reagent to reveal the

grain structure. Grains in the DMLS material were very large and irregularly shaped as compared to the

regular austenite grains found in the sheet. Figure 10 shows microscope imagery of the grains in the as-

received DMLS material. Note that in the lower middle region, some scan tracks can still be faintly made

out. Average grain sizes were generally much larger than those found in the sheet material, with

average effective grain diameters of 64 µm in the DMLS material and 14 µm in the sheet as-received.

Figure 11 shows the austenite grains in the sheet material for comparison; these grains are much more

typical of the expected appearance of austenite. Notably, the grains in the DMLS material traverse scan

tracks, suggesting that crystallization is taking place across the boundary of the weld trough.

Some DMLS samples were annealed to see if heat treatment established a more regular grain

structure. An annealing cycle at 1100° was applied per the ASM Handbook, but did not seem to have an

effect on the average grain size. Figure 12 shows the DMLS grain structure after rolling to the 20%

thickness reduction level. Rolling reduced the average effective grain diameter approximately 10% from

the as-received material, but grains were still large and irregularly shaped; follow-up work has revealed

that the grain structure may be significantly different from traditional austenite due to non-equilibrium

heating and cooling in the laser-sintering process. Annealing the material had little effect on the as-

received DMLS grain size and shape. Vickers hardness tests (HV50) were also run on both the sheet and

DMLS material in the as-received condition; the DMLS material had an average hardness of

approximately 220 kg/mm2 while the sheet material averaged 190 kg/mm2.

Figure 10: DMLS 316L grain structure as-received. Scale at lower left indicates 100 μm.

Figure 11: SEM image of austenite grain structure in 316L sheet as received

Figure 12: Grain structure of DMLS 316L with 20% thickness reduction by rolling. Scale at lower left indicates 100 μm.

Corrosion of these samples, as will be discussed later, illuminated an important aspect of the

microstructure of the DMLS material: it appears likely that the structures observed are not necessarily

austenitic in nature. Figure 13 shows SEM imagery of the bottom of a corrosion pit in the DMLS material.

The small structures shown do not appear to be similar to those found in rolled sheet of the same bulk

chemical composition; notably, these features disappear after an annealing cycle, after which, the

material’s corrosion response appears more regular. General understanding of the establishment of

regular microstructure in traditionally manufactured metals is predicated on equilibrium heating and

cooling, which is typical in processes such as casting, rolling, and drawing. DMLS, however, involves

instantaneous heating of small particles that have infinitesimal thermal mass, followed very shortly by

instantaneous welding of those particles to an effectively infinite heat sink. This appears to result in

microstructures that are as-yet uncharacterized. Continued work is being done to model these heating

and cooling processes in order to characterize the fundamental structures that are created through

laser/powder additive metal processes.

Figure 13: Scan tracks and leftover structure in corrosion pit of DMLS 316L stainless steel

4.2 Rolling

The rolling process went generally as predicted by the modeling. The rolling schedule induced

roughly the level of desired thickness change, although the transverse samples had slightly less thickness

reduction over the course of the rolling schedule than the longitudinal and sheet samples. The initial

surface finish of the DMLS samples was remarkably rough, but the first rolling pass flattened the surface

significantly. Figure 14 shows a three dimensional optical profilometry map of the as-received DMLS

surface. Note that spherical beads of melted powder dominate the source of roughness. This may have

kept the thickness strain localized to the surface for the first rolling pass, but as will be shown, there was

significant evidence that the strain penetrated through the thickness of all three materials. Figure 15

shows microscope imagery of the as-received DMLS surface, while Figure 16 shows that of the rolled

surface.

Figure 14: Optical profilometry of as-received DMLS surface, showing total roughness of 120 µm

Figure 15: Surface of 316L DMLS material as manufactured

Figure 16: Surface of 316L DMLS material after rolling to 20% thickness reduction

4.3 Tensile Testing

In tensile testing, the materials generally performed as expected. 20 samples of each DMLS material

treatment were tested, as well as samples of the same geometry from a single sheet of 316L material.

Initial tensile performance was comparable between the two build directions, with the transverse

direction having much wider variability between specimens than the longitudinal direction. Both the

longitudinal direction DMLS material and the sheet material were extremely consistent in their stress-

strain behavior. The as-received DMLS material had a relatively high yield strength (approximately 70 ±1

ksi in the longitudinal direction and 75 ±6 ksi in the transverse direction) as compared to the annealed

sheet material (47 ksi).

The DMLS material, as was found in the initial work leading to this study, was far less ductile, with

less than half the strain before failure than the standard sheet. The failure surfaces of the DMLS material

were notably brittle in their appearance as compared to the sheet material at low levels of thickness

reduction, as shown in Figure 17. At the 15%-20% rolling steps, the fracture surfaces of the longitudinal

DMLS samples appeared the same as the standard sheet samples, but the surfaces of the transverse

samples appeared unaffected by the thickness reduction.

Figure 17: Fracture surfaces for as-received DMLS (left) and standard sheet (right) tensile samples

Tensile tests of specimens rolled to each successive rolling step were superimposed as shown in

Figures 17, 18, and 19. This technique allows calculation of the amount of strain hardening exhibited by

each specimen in its worked state. This total strain, the amount of true strain introduced into the curve

to match it with the strain hardening curve of the unworked sample, can then be compared to the

apparent strain realized in shape change in order to assess the level of redundant strain being put into

the material. This process was applied to all three materials, so as to compare the performance of the

DMLS material to that of the sheet, using the sheet as a control.

Figure 18: Superimposed true stress-true strain curve for 316L sheet material for successive rolling steps

Figure 19: Superimposed true stress-true strain curve for 316L DMLS material in the longitudinal (Z) direction for successive rolling steps

Figure 20: Superimposed true stress-true strain curve for 316L DMLS material in the transverse (XY) direction for successive rolling steps

As mentioned previously, it can be seen that the results were much more consistent for the

longitudinally oriented DMLS samples than for the transverse. The total strain values obtained from this

analysis were plotted against the apparent strain values in Figure 21. In this plot, the reference points

for the sheet material are represented by the red crosses, while the blue line represents the minimum

level of work necessary to introduce a given amount of apparent strain into the material. Points below

this line exhibited lower amounts of strain hardening than was indicated by the actual shape change.

Notably, the transverse specimens (gray triangles) do so; this is likely due to internal porosity and/or

inhomogeneity in the material. The general plasticity theory from which the modeling in this case

derives is generally predicated on both a material continuum and homogeneity. It appears that the

material inconsistencies in the transverse specimens were eliminated in the first rolling step, at which

point the material began to work harden at a similar rate to the longitudinal specimen.

Figure 21: Apparent true strain as measured by actual thickness reduction vs. total true strain reflected in superposition of strain-hardening curves for various levels of thickness reduction and material treatments

As shown by the sheet performance as compared to the ideal strain line, redundant strain exhibited

by the sheet material was relatively low. The model discussed previously predicted slightly lower

redundant strain than was evident in the sheet rolling results, shown with blue diagonal crosses. Since

the theory generally considers geometry alone, it is assumed that since the DMLS geometry was the

same as that of the sheet, the level of redundant strain introduced into the DMLS samples should have

been relatively similar. In fact, both DMLS samples exhibited a higher rate of work hardening than could

be explained by the redundant strain model, with the longitudinal samples (black circles) showing

approximately four times the increase in total true strain as evidenced by strain hardening over the ideal

work-hardening line as compared to the sheet material. This, coupled with much lower levels of

expected redundant strain, suggests that a strengthening mechanism not explained through the existing

redundant strain model was involved.

DMLS specimens exhibited less than ideal strain in the first rolling step, indicated by the horizontal

offset in their respective fit lines. This is likely due to densification of internal pores and localized

deformation of surface imperfections in the initial rolling step. The two DMLS treatments had similar

rates of increase, indicating similar relationships between ideal strain behavior and the actual strain

hardening that occurred. These differed significantly from the rates of increase indicated by the

conventional material and ideal lines, indicating a phenomenon unrelated to the deformation zone

geometry alone. If the deviation of the data from the redundant strain model was due to non-constant

flow stress, it would be expected that the DMLS samples would have deviated less from the ideal model

than the sheet material, as there was less work hardening in the DMLS material.

Microhardness testing was conducted across the cross-section of rolled samples of DMLS and sheet

to characterize the gradient of strain hardening across the sample cross-sections. Figure 22 shows the

Vickers hardness across the cross-section of sheet (red crosses) and DMLS (black circles) samples in the

unrolled and rolled conditions. The results appear to show two key phenomena: first, there appears to

be a significant gradient in the DMLS sample that is not present in the sheet sample, with the outside

edges of the DMLS sample appearing to have experienced more strain hardening than those of the sheet

sample. This appears to be verified by the significant changes in aspect ratio of the scan tracks as shown

previously in Figure 2. Second, the DMLS sample appears to have experienced less hardening than the

sheet sample on average, which correlates with the amount of strain hardening demonstrated in Figure

18 and Figure 19, respectively. The results show an increase in hardness with respect to a 20% reduction

for both materials. The conventional material experienced greater and more uniform hardening than

the DMLS material, consistent with the amount of hardening shown in the tensile tests. The DMLS

material exhibits about a 60 HV gradient from the center to the surface, as would be expected in the

presence of redundant strain.

Figure 22: Vickers hardness from microhardness testing of DMLS and standard sheet 316L after rolling to 20% thickness reduction

Figure 23 shows the density of the DMLS samples at each level of rolling, with the blue line

indicating the full density of the sheet material at 8.04 g/cm3. Notably, the transverse specimens

exhibited very high density even in the as-received state, and appeared not to change significantly over

the course of the rolling schedule. The longitudinal specimens exhibited some measured densification

through rolling in the first step as previously indicated by the offset in Figure 21, but with relatively high

variation among the subsequent steps, indicating a relatively weak relationship beyond the first (~5%)

strain level. In order to determine whether the densification had a direct correlation with the non-ideal

strain, Figure 24 plots the difference between the apparent strain and total strain vs. the absolute

density of the specimens. There is no apparent relationship among them, indicating that the level of

density itself does not explain the apparent strengthening of the samples. This says nothing about the

orientation of the pores or inhomogeneities in the material, which could not be easily measured in this

study. The data strongly suggest that the directionality of the DMLS lay-up with respect to that of the

induced plastic strain had a significant effect.

Figure 23: DMLS specimen density at different levels of thickness reduction by rolling

Figure 24: Difference between total strain and apparent strain in longitudinal (Z) DMLS specimens vs. post-deformation specimen density

Figure 25 shows the effect rolling deformation had on the strain ratio, R, of the specimens. Notably,

the sheet material, indicated by red crosses, showed more initial tensile anisotropy than the DMLS

specimens did. As the sheet material was rolled, its R value decreased significantly. The longitudinal

specimen’s behavior mirrored that of the sheet specimen, being reduced by roughly the same

proportion for each rolling step. The transverse specimen, however, deviated from this behavior, with

its R value climbing significantly, similar to the inverse of the longitudinal strain ratio.

Figure 25: Strain ratio at different levels of thickness reduction by rolling for 316L sheet and DMLS material in longitudinal (Z) and transverse (XY) print directions

4.4 Fatigue

As mentioned in section 4.2, the surface finish of the DMLS samples was significantly improved

through the rolling process. The standard sheet in the as-received condition generally had a surface

roughness less than Ra 16 μin, while the as-received DMLS material had roughness values between Ra

2000 and 4000 μin. In order to prevent this surface roughness from obscuring the effects of plastic

deformation on fatigue, samples were polished to as consistent a surface finish as possible. Some

materials exhibit a significant decrease in fatigue performance when the surface layer is sufficiently

removed to ensure a smooth surface finish because of the elimination of a thin, hard layer or residual

compressive stresses on the surface.

Polishing the DMLS samples removed approximately 5% of depth from each side, which may have

had a slight effect, but it is unlikely to have been a significant detraction in fatigue performance as

compared to the surface finish. Optical microscope imagery of the resultant surface is shown in Figure

26. As can be seen, the polishing process was limited by internal pores revealed as material was

removed from the surface. The surface between the pores was polished to a surface finish of Ra 16 μin

so as to be consistent with the sheet to which it would be compared. Rolled DMLS samples were able to

be machined back to the ASTM B593 geometry, but rolled sheet samples were not, notably because

they increased in length much more than the DMLS specimens. This seemed to confirm deviation from

assumptions of a material continuum as previously discussed. Investigation is still ongoing to

characterize a new geometry so as to produce corollary S-N data for the rolled sheet samples, as will be

published in a journal paper resulting from this work for a full comparison. Elastic modulus values were

also significantly different, averaging 20x106 psi for DMLS samples of both treatments, while sheet

material averaged 25.8x106 psi.

Figure 26: Surface of 316L DMLS material after rolling to 20% true thickness strain and polishing to average surface finish of Ra 16 μin, with exposed pores visible.

Figure 27 shows the S-N data for the different material treatments. The data were well-behaved,

showing relatively low levels of scatter compared to what is often seen in fatigue testing. As expected,

the fatigue performance of DMLS 316L in the unrolled condition (black circles) was significantly lower

than that of traditional sheet material (red crosses) in the unrolled condition. Polishing the sheet

samples (red diagonal crosses) seemed to significantly increase the endurance limit, but did not

otherwise seem to have an effect. Failures were not detected for sheet samples past approximately 1.5

million cycles. DMLS samples continued to fail well past the 106 cycle level, exhibiting no detectable

endurance limit up to 5 million cycles. Rolled DMLS material performance was indistinguishable from as-

received DMLS material according to the S-N data.

Figure 27: S-N data for DMLS and sheet 316L stainless steel in various material treatments

Rolled DMLS sample stresses were corrected for the fact that the tapered geometry of the sample

led to some taper in the specimens. The stresses plotted are the maximum nominal stress in the sample

based on assumptions of elasticity. In the rolled samples, there was a slight gradient of stress through

the gage section, and some failures occurred at a point in the gage section where stresses were not a

nominal maximum. In fact, all points on the graph represent the nominal stress levels as computed by

the elastic deflection, which is both standard practice, and a slight underestimation of actual maximum

stresses. Figure 28 shows the stress concentrations present in the B593 specimen geometry when

deflected as described in the test. The stress concentrations along the edges total a static stress

concentration factor of approximately 1.2. For relatively notch-insensitive materials such as annealed

316L stainless steel, this is unlikely to have a significant effect on S-N performance, but for notch-

sensitive materials, such as cold-worked 316L, it may have an impact on resulting S-N data.

Figure 28: FEA simulation of stresses in ASTM B593 specimen gage section as deflected in Krouse-type fatigue bending test

The fracture surfaces resulting from the fatigue tests were remarkably different. A typical fracture

surface from the 316L as-received sheet is shown in Figure 29. While difficult to picture from the image,

this surface is relatively straight, as shown in Figure 30. The line down the center of the specimen is the

neutral axis, which is manifest as a sort of ‘shelf’ where cracks propagating from slightly different

locations on each surface meet just before fracture. Polished and unrolled DMLS, however, had an

extremely irregular fracture surface that seemed to be dominated not by sharp crack growth, but

instead by rounded paths that left powder artifacts from the printing processes intact. Figure 31 shows

optical microscopy of the fracture surface of an unrolled DMLS specimen, with the same characteristic

spherical beads as seen on the surface in Figure 15 and internally in Figure 32. Rolled DMLS samples, as

shown in Figure 33 did not have the same fracture surface appearance as the unrolled DMLS samples,

but still did not exhibit surfaces commensurate with normal crack growth in continuous material.

Figure 29: 316L sheet fatigue fracture surface, showing neutral axis

Figure 30: 316L sheet fatigue sample with characteristic crack just before final fracture

Figure 31: Unrolled DMLS fatigue fracture surface, showing internal spherical print artifacts intact.

Figure 32: Internal pore in unrolled DMLS fatigue sample, showing spherical beads present on the internal surface of the void

Figure 33: Rolled DMLS fatigue fracture surface

Though the fatigue data did not reflect a significant change in performance between the unrolled

and rolled DMLS, mounted and polished cross-sections revealed significant differences in the two

material treatments. Figure 34 shows a significantly porous cross-section of an unrolled DMLS sample.

Scratches were difficult to avoid due to media getting trapped in the pores. Scan tracks can be made

out, and appear to be consistent in width and aspect ratio, being longer across the thickness of the

sample than along the width. Figure 35 shows a cross section from a rolled DMLS sample, reflecting

much smaller pores and what appears to be a significantly deformed layer of scan tracks on the surface.

Overall density of the samples was similar, so the difference in pore size may be simply due to having

taken a separate cross-section for the two images, but if the scan tracks deform with the material, this

may be evidence that the compressive stresses did indeed change the pore geometry and size.

Nonetheless, likely due to the process being done cold, artifacts of the DMLS process are still evident in

the microstructure of the material, and still appear to dominate its fatigue performance.

Figure 34:Cross-section of unrolled DMLS fatigue sample showing significant internal porosity and scan tracks

Figure 35: Cross-section of rolled DMLS fatigue sample showing reduced internal porosity and persistence of scan tracks

4.5 Corrosion

Corrosion results were quite similar between the sheet and rolled DMLS material at first glance.

Both materials had highly localized pits, but relatively little other surface reaction to the ferric chloride.

Noting that the samples were the same size and polished to the same surface finish, a reduction in

density, characterized in Figure 36 as porosity, was significantly higher for the DMLS material. Rolled

DMLS sample performance was improved to slightly better than that of as-received sheet material. An

annealing treatment was added to the test battery, and was found to have a significant effect on the

unrolled DMLS samples, but otherwise did not. The primary driver behind these differences is still under

study, and further experiments have been planned.

Figure 36: Relative porosity of different treatments of 316L stainless steel sheet and DMLS material in un-annealed and annealed states before and after pitting corrosion with a 10% FeCl3 solution

5. Conclusions and Recommendations

This work set out to test the following hypotheses:

Tensile properties, including yield strength, ultimate tensile strength, and anisotropy of DMLS

SS316L are affected by compressive plastic deformation beyond a level commensurate with

work hardening of the material

High-cycle fatigue life of DMLS SS316L is affected by compressive plastic deformation beyond

what is expected due to simple work hardening of the material

Pitting corrosion resistance of DMLS SS316L is improved by compressive plastic deformation as

compared to samples of the same material with the same surface properties

Tensile properties of DMLS material did appear to be affected by compressive plastic deformation.

Directionality was a factor, and the overall level of densification appears not to have a significant effect,

so the mechanism by which the material strengthening and anisotropy are affected is as yet unknown.

Deformation alone did not break down scan tracks, although there was evidence that it did deform

0

1

2

3

4

5

6

7

8

9

AsReceived

HeatTreated

AsReceived

HeatTreated

AsReceived

HeatTreated

As Rolled DMLS DMLS + Cold Work

Poro

sity (

%) Before Test

After Test

them. Grain structure did refine under deformation, but was still relatively coarse and highly irregular in

shape. SEM imagery and EDS analysis of corroded DMLS samples suggest that the microstructures

present in that material differ fundamentally from those found in samples that experienced equilibrium

cooling. This experiment was run in such a way as to prevent obfuscation of the effects of plastic

deformation from the effects of heat, which likely limited the results in terms of alteration of grain

structure. The combination of heat and deformation would likely yield different results.

DMLS specimens exhibited greater increases in total true strain as measured by superposition of

stress-strain curves than sheet specimens in the same rolling schedule. DMLS specimens only showed a

change in density in the first rolling step, evidenced by a delayed onset of strain hardening as compared

to the thickness strain and stress-strain curve of the as-received material. There was otherwise no

observable relationship between DMLS material densification and deviation of the total strain from the

ideal strain line. Deformation did not break down scan tracks, although there was evidence that it did

deform them. Grain structure in DMLS samples was larger and more irregular than that in the

conventional material, even after deformation.

Microhardness testing showed more overall strain hardening from cold-rolling in the conventional

sheet than in the DMLS specimens. Conventional 316L sheet exhibited essentially uniform hardness

across the specimen thickness indicating minimal redundant strain up to the 20% cold rolling, consistent

with tensile test results. There was a measurable hardness gradient across the thickness of the DMLS

material, with a softer interior and harder surface, indicative of a strain gradient. DMLS tensile test

results are also consistent with a higher degree of redundant work, which was not predicted by a

classical analysis of the deformation zone geometry. The deviation of both DMLS configurations from

the redundant work model was significant.

High cycle fatigue performance was unaffected by compressive plastic deformation. Fracture

surfaces were irregular as compared to those found in traditional rolled sheet, likely a result of internal

material boundaries developed in the laser sintering process. Further study with additional treatments

of elevated temperature during deformation and directionality are worth pursuing, so as to better

understand the fundamentals behind the behavior. Further, the cause of the lowered elasticity is worth

studying not just from the standpoint of fatigue testing, but also for designs that rely on stiffness as

opposed to strength for stability. If deformation processes could be used to improve elastic modulus,

they could be instrumental in optimization of lightweight structures to avoid buckling.

According to the limited scope pursued in this study, pitting corrosion resistance of DMLS 316L

stainless steel was significantly improved by compressive plastic deformation, in terms of mass loss. The

mechanism by which this mass loss is occurring is not well understood, and is currently under study with

a broader test battery. A better understanding of the underlying fundamental mechanism will allow

tailoring of the process to consistently provide enhanced material performance in a comprehensive

post-process treatment.

In short, it appears from these results that forging and related deformation processes could serve as

an important post-process to maximize the tensile and corrosion performance of DMLS metals. More

characterization would be needed to understand the mechanisms by which the observed phenomena

are occurring in order to optimize an actual post-process. Results from this work suggest that the scan

tracks in the additive print layup have significant directional effects on the resulting material

performance that could be modeled and manipulated for process optimization. This work has provided

both the experimental experience and the impetus for construction of basic testing infrastructure and

procedure to more easily explore these fundamentals with a goal of actual process implementation.

These results also suggest that phenomena involved in the strain-hardening behavior of DMLS material

that go outside traditional models of metalforming. Further work to characterize the nature of scan

tracks and to model their effects on plasticity in laser-sintered parts is necessary to understand the

possibilities for improving the mechanical integrity of parts made through those processes.

6. Bibliography [1] N. Kurgan, “Effect of porosity and density on the mechanical and microstructural properties of

sintered 316L stainless steel implant materials,” Mater. Des., vol. 55, pp. 235–241, Mar. 2014.

[2] L. Ladani, J. Razmi, and S. F. Choudhury, “Mechanical Anisotropy and Strain Rate Dependency

Behavior of Ti6Al4V Produced Using E-Beam Additive Fabrication,” J. Eng. Mater. Technol.-Trans. Asme,

vol. 136, no. 3, p. 031006, Jul. 2014.

[3] Y. Wang, J. Bergstrom, and C. Burman, “Four-point bending fatigue behaviour of an iron-based

laser sintered material,” Int. J. Fatigue, vol. 28, no. 12, pp. 1705–1715, Dec. 2006.

[4] N. Chawla and X. Deng, “Microstructure and mechanical behavior of porous sintered steels,”

Mater. Sci. Eng. -Struct. Mater. Prop. Microstruct. Process., vol. 390, no. 1–2, pp. 98–112, Jan. 2005.

[5] K. Christian and R. German, “Relation Between Pore Structure and Fatigue Behavior in Sintered

Iron-Copper-Carbon,” Int. J. Powder Metall., vol. 31, no. 1, pp. 51–61, Jan. 1995.

[6] D. Gu and Y. Shen, “Processing conditions and microstructural features of porous 316L stainless

steel components by DMLS,” Appl. Surf. Sci., vol. 255, no. 5, pp. 1880–1887, Dec. 2008.

[7] Y. Y. Du, Y. S. Shi, and Q. S. Wei, “Forming Simulation and Experimental Verification of Combined

Formation of Selective Laser Sintering and Cold Isostatic Pressing,” J. Mater. Eng. Perform., vol. 20, no. 2,

pp. 185–190, Mar. 2011.

[8] S. Das, M. Wohlert, J. J. Beaman, and D. L. Bourell, “Producing metal parts with selective laser

sintering hot isostatic pressing,” Jom-J. Miner. Met. Mater. Soc., vol. 50, no. 12, pp. 17–20, Dec. 1998.

[9] J. H. Liu, Y. S. Shi, Z. L. Lu, and S. H. Huang, “Manufacturing near dense metal parts via indirect

selective laser sintering combined with isostatic pressing,” Appl. Phys. -Mater. Sci. Process., vol. 89, no.

3, pp. 743–748, Nov. 2007.

[10] Z. Wang, Y. Shi, R. Li, Q. Wei, and J. Liu, “Manufacturing AISI316L Components via Selective Laser

Melting Coupled with Hot Isostatic Pressing,” in Advanced Material Science and Technology, Pts 1 and 2,

Stafa-Zurich, 2011, vol. 675–677, pp. 853–856.

[11] K. T. Kim and Y. C. Jeon, “Densification behavior of 316L stainless steel powder under high

temperature,” Mater. Sci. Eng. -Struct. Mater. Prop. Microstruct. Process., vol. 245, no. 1, pp. 64–71,

Apr. 1998.

[12] W.-S. Lee, C.-F. Lin, and T.-J. Liu, “Impact and fracture response of sintered 316L stainless steel

subjected to high strain rate loading,” Mater. Charact., vol. 58, no. 4, pp. 363–370, Apr. 2007.

[13] W. Morton, S. Green, A. E. W. Rennie, and T. N. Abram, “Surface finishing techniques for SLM

manufactured stainless steel 316L components,” Innov. Dev. Virtual Phys. Prototyp., pp. 503–509, 2012.

[14] W. F. Hosford and R. M. Caddell, Metal Forming: Mechanics and Metallurgy, 7th ed. Cambridge

University Press, 2011.